Will GPUs with 4GB VRAM age poorly?

Fury cards seem to age poorly in any memory intensive game nowadays, the observation is based on two things:

1-Massive fps drops on the Fury cards compared to competition, when maximum visual settings are enabled.
2-390 cards having close (equal or better) fps than Fury cards due to having 8GB of RAM.


The games are:

  • DOOM
    Game insists on 5GB or more for the Nightmare texture and shadow settings
 
Last edited:
Fury cards seem to age poorly in any memory intensive game nowadays, the observation is based on two things:

1-Massive fps drops on the Fury cards compared to competition, when maximum visual settings are enabled.
2-390 cards having equal or better fps than Fury cards due to having 8GB of RAM.

The games are:
Call Of Duty Black Ops 3
DOOM
Mirror's Edge Catalyst
Rainbow Six Siege Only @4K
Rise Of The Tomb Raider (?) (need more data on this one, but it's a memory hog nevertheless)

Any thoughts?


I'm not sure. Do you think AMD ever had any concerns about whether the Fury's 4GB of VRAM was appropriate for a flagship card released in 2015?


AMD-Radeon-R9-390_Official_3-635x357.jpg
 
Well this is physics 101...less memory will always be worst than more memory...but in the case of the Fury the memory BW sort of elevates the need for more VRAM in most cases (as seen with the Fury X's 4K performance which is for the most part always faster than a stock 980 Ti). Performance gets trashed once "ultra" textures are enabled..and like the linked PC Gamer article says ""Requires" 5GB of VRAM but only looks slightly better". A 8Gb HMB Fury/Nano/X would have definitely been more "futur proof" though..
 
Well this is physics 101...less memory will always be worst than more memory...but in the case of the Fury the memory BW sort of elevates the need for more VRAM in most cases (as seen with the Fury X's 4K performance which is for the most part always faster than a stock 980 Ti).
This is a good time to bring out the "correlation doesn't mean causation" warning, and the story of a blind men describing an elephant.

Fiji performs better with higher resolutions, but I'm not at all convinced that this is due to increased BW. The relative improvements are minor, a few percentage points more or less.
One could explain the performance in other ways.

E.g. with higher resolutions, the amount of triangles per pixels is smaller than with lower resolutions. This may reduce the pressure on some internal geometry related bottleneck.

Not saying that this is the reason, it's probably a mix of many smaller factors (with higher BW just one of them.)
 
Well this is physics 101...less memory will always be worst than more memory...but in the case of the Fury the memory BW sort of elevates the need for more VRAM in most cases[...]
I think this is the opposite of the point you were trying to make :D
 
Last year when the Furies were premiered, AMD stated they would spend resources on optimizing driver memory management to alleviate RAM pressure due to only having 4GB on-board.

I wonder how much effort they actually spent in that regard.
 
Does the same apply to other cards with lesser amounts of ram, like the 4GB 980s ?
 
Yep, a flagship card with only 4GB memory looks pretty limited in 2015. I'm pretty sure I said as much, before it launched, since we knew that's all HBM1 would allow for.

Some of these games were tweaked to favour NVidia's just-launched 8GB 1080, weren't they?
 
Last year when the Furies were premiered, AMD stated they would spend resources on optimizing driver memory management to alleviate RAM pressure due to only having 4GB on-board.

I wonder how much effort they actually spent in that regard.

Some games do use less VRAM on the Fury but the same also applies to GTX too in other games. I guess that it depends on which GPU architecture the game/engine was build up on/optimized for.

The Division for example uses way less VRAM on the Fury:
http://imgur.com/a/v3jUh
 
Last edited:
The Division for example uses way less VRAM on the Fury:
Yeah, well, you don't really want a few special cases, you want a general solution (as much as possible anyway), because maybe you don't play The Division.

Someone mentioned DOOM where the Fury took a header, fps-wise. The Division's optimization either doesn't work for DOOM, or it is actually inherent to the Division, not AMD's driver... :p
 
Some games do use less VRAM on the Fury but the same also applies to GTX too in other games. I guess that it depends on which GPU architecture the game/engine was build up on/optimized for.

The Division for example uses way less VRAM on the Fury:
http://imgur.com/a/v3jUh
Memory usage is a lot more complex then people generally realise. There's render target compression for both color and z buffers (and for which I feel it needs to be immediately said that it doesn't save on total space required for memory buffer). They may require a separate decode step (more memory) before texturing from. There's resource alignment. Then there are dynamic buffers (discard). There's rendering stuff ahead...
So creating a 4096x4096 DXGI_FORMAT_R8G8B8A8_UNORM surface in the end might not cost you just 64MB...
 
With a 4GB card being "fully used" I suspect if you classified the gaming memory usage into: read-only textures and all other uses, you'd get an 80:20 split.

Apparently some games that use streaming-world techniques don't cause a problem for cards that only have 4GB. Though the act of streaming data to the GPU can cause stutter.
 
There seems to be something wrong with those results in Fiji cards. Guru3D has much better performance on them:
http://www.guru3d.com/articles_page..._graphics_performance_benchmark_review,8.html
The 290X gets 10% less performance than the 390X and all Fury cards beat the 390X very easily.

To be honest, I'm surprised at how AMD's OpenGL driver is making their cards look so competitive in this game.
Maybe the memory optimizations didn't (and won't) reach the OpenGL driver at all?
Regardless, both id software and AMD have been promising Vulkan as the go-to solution for Doom in AMD cards (even though it's incredibly, embarrassingly late), which makes sense since the IHV pretty much developed most of the API.

From that same link:
oEzigCQ.png


If you're talking about the "Hyper" mode, it looks like the game itself is artificially preventing cards with less than 6GB VRAM from running it.

Maybe the only game where memory amount may be the problem (though I can't find any other comparison in the internet?).
Regardless, meh gameworks-ridden title..


Rise Of The Tomb Raider (more so @4K) (need more data on this one, but it's a memory hog nevertheless)
Can't read Polish, but aren't those results pre-AMD ready drivers?

The 4GB don't seem to be a limitation at 4K here:
http://www.techspot.com/review/1128-rise-of-the-tomb-raider-benchmarks/page4.html

And with the latest DX12 patch, the Fury X gets better performance than the 6GB 980Ti at 4K:
c55Dm1Y.jpg




Regardless, I have no doubts those 4GB will eventually become a hindrance for 4K and up.
It just wasn't a limitation for 2015, isn't a meaningful limitation for 2016 and it probably won't be for most of 2017 either.
Come 2018, few people will care. By then, I doubt the ones still clinging to a Fury will either be playing in a 4K monitor or expecting playable performance with maxed out settings.
 
There seems to be something wrong with those results in Fiji cards. Guru3D has much better performance on them:
http://www.guru3d.com/articles_page..._graphics_performance_benchmark_review,8.html
The 290X gets 10% less performance than the 390X and all Fury cards beat the 390X very easily.
These results are taken with one quality settings down, which is Shadow Maps (set at High in Guru3d test), The maximum setting is (Extra), it consumes large amount of memory, in which Fury cards choke.
To be honest, I'm surprised at how AMD's OpenGL driver is making their cards look so competitive in this game.
Maybe the memory optimizations didn't (and won't) reach the OpenGL driver at all?
Maybe, regardless, it still chokes on the maximum texture setting. maybe Vulkan will be the same too.
From that same link:
If you're talking about the "Hyper" mode, it looks like the game itself is artificially preventing cards with less than 6GB VRAM from running it.
It doesn't, the game runs but it's a stuttering mess, so they didn't bother testing. Also Fury cards take a huge hit even at Ultra (the direct step down from Hyper).
Maybe the only game where memory amount may be the problem (though I can't find any other comparison in the internet?).
Regardless, meh gameworks-ridden title..
Almost all AAA titles are. So the point is? In fact this game is very light on GameWorks effects.

And with the latest DX12 patch, the Fury X gets better performance than the 6GB 980Ti at 4K:
Seems the new patch changed things for the better with AMD. Still more data is needed with this game, certain sections are light on memory. Techreport testings in the GTX 1080 review are based on the latest drivers from both sides, and it shows Fury cards taking massive hit in DX11.

Regardless, I have no doubts those 4GB will eventually become a hindrance for 4K and up.
It is a hindrance NOW even at 1080p.
 
Last edited:
Back
Top